Scholarlink

<p>The thermally induced free vibration behavior of curved laminated composite and sandwich beams is investigated using a rigorously coupled computational–data-driven framework. Curved geometries introduce membrane–bending coupling and anisotropic interactions that significantly complicate dynamic response prediction, particularly under thermal loading where material degradation alters stiffness characteristics. To address this, a higher-order zigzag theory (HOZT)-based finite element formulation is developed to accurately capture interlaminar kinematics and transverse shear effects without ad hoc corrections. The numerical model is validated against literature benchmarks and ABAQUS simulations, demonstrating high fidelity. A large-scale parametric dataset (10,000 samples) is generated using Sobol sequence sampling across geometric, material, layup, boundary, and thermal variables. A hybrid physics-informed artificial neural network–support vector machine (ANN–SVM) surrogate is then constructed, embedding variational bounds, sensitivity constraints, and uncertainty quantification within the learning process. This is not just curve-fitting, the model is explicitly regularized by mechanics. Results show a consistent reduction in natural frequencies with increasing temperature due to stiffness degradation, while curvature and ply orientation induce nontrivial modal coupling effects. The proposed hybrid model outperforms standalone approaches, achieving R<sup>2</sup> ≈ 0.98 (training) and 0.96 (testing), with unbiased residuals and well-calibrated prediction intervals.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Thermally-induced free vibration behavior of curved laminated composite and sandwich beams using high-fidelity hybridized physics-informed neural network-support vector machine learning algorithm

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Ashish,
  • S. Anbukumar

摘要

The thermally induced free vibration behavior of curved laminated composite and sandwich beams is investigated using a rigorously coupled computational–data-driven framework. Curved geometries introduce membrane–bending coupling and anisotropic interactions that significantly complicate dynamic response prediction, particularly under thermal loading where material degradation alters stiffness characteristics. To address this, a higher-order zigzag theory (HOZT)-based finite element formulation is developed to accurately capture interlaminar kinematics and transverse shear effects without ad hoc corrections. The numerical model is validated against literature benchmarks and ABAQUS simulations, demonstrating high fidelity. A large-scale parametric dataset (10,000 samples) is generated using Sobol sequence sampling across geometric, material, layup, boundary, and thermal variables. A hybrid physics-informed artificial neural network–support vector machine (ANN–SVM) surrogate is then constructed, embedding variational bounds, sensitivity constraints, and uncertainty quantification within the learning process. This is not just curve-fitting, the model is explicitly regularized by mechanics. Results show a consistent reduction in natural frequencies with increasing temperature due to stiffness degradation, while curvature and ply orientation induce nontrivial modal coupling effects. The proposed hybrid model outperforms standalone approaches, achieving R2 ≈ 0.98 (training) and 0.96 (testing), with unbiased residuals and well-calibrated prediction intervals.